pith. sign in

arxiv: 2606.07416 · v2 · pith:OKWVMT4Hnew · submitted 2026-06-05 · 💻 cs.LG

Video-Based Prediction of In-Flight Particle Characteristics in Atmospheric Plasma Spraying

Abhijeet Praveen , Sareh Soleimani , Cormac Cureton , Aman Sidhu , Kintak Raymond Yu , Cristian Cojocaru , Narges Armanfard This is my paper

Pith reviewed 2026-06-27 22:45 UTC · model grok-4.3

classification 💻 cs.LG
keywords atmospheric plasma sprayingin-flight particleshigh-speed videotemperature predictionvelocity predictionTabPFNCNNnon-invasive monitoring
0
0 comments X

The pith

High-speed video of the plasma plume predicts in-flight particle temperature and velocity in APS using ML models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether observations from high-speed video of the plasma plume can estimate the temperature and velocity of particles in flight during atmospheric plasma spraying. Direct measurement of these characteristics is difficult during operation, so the work develops three video-derived feature sets and applies TabPFN, CNNs, and classical regressors to data from 126 labeled recordings across 63 spray runs under grouped cross-validation. If the predictions hold, they would support non-invasive, real-time monitoring that could improve control over final coating properties. A sympathetic reader would care because better process visibility addresses a practical barrier in industrial coating production without adding sensors into the harsh plume environment.

Core claim

Video-derived plume information provides a promising and scalable foundation for non-invasive APS diagnostics and real-time process monitoring, with TabPFN reaching R²=0.86 for temperature on combined features and pretrained CNNs reaching R²=0.90 for temperature and 0.82 for velocity on raw frames.

What carries the argument

High-speed video recordings of the plasma plume converted into engineered feature representations or used as raw frames, processed by TabPFN or pretrained CNN regression heads to predict particle temperature and velocity.

Load-bearing premise

The 126 labeled video recordings from only 63 APS spray runs are representative of the full range of operating conditions, particle sizes, and plume variations in production.

What would settle it

Substantially lower R² scores than 0.82–0.90 when the same models are tested on video from new spray runs, different equipment, or operating conditions outside the original 63 runs.

Figures

Figures reproduced from arXiv: 2606.07416 by Abhijeet Praveen, Aman Sidhu, Cormac Cureton, Cristian Cojocaru, Kintak Raymond Yu, Narges Armanfard, Sareh Soleimani.

Figure 1
Figure 1. Figure 1: Representative first, second and third temporal frames from APS plume videos (from left to right), illustrating the visual information available in the spray recordings. The overlays show the segmented plume core, the region of interest (ROI) used for fea￾ture extraction, and the centroid trajectory used to characterize plume motion. These visual cues form the basis for the engineered video feature represe… view at source ↗
Figure 2
Figure 2. Figure 2: Distribution of representative video-derived plume features across APS spray conditions. Each grouped boxplot compares the normalized distributions of selected intensity-based and geometric descriptors under different gas flow configurations (Ar￾gon/Nitrogen/Hydrogen in LPM). 4 Experiments This section describes the modeling approaches used in this study and their evaluation. We first present the models ap… view at source ↗
read the original abstract

Atmospheric plasma spraying (APS) is a widely used coating process in which in-flight particle temperature and velocity strongly influence coating quality. However, these particle characteristics are difficult to monitor continuously during operation, motivating the development of non-invasive data-driven diagnostic methods. In this work, we investigate the predictive potential of high-speed video observations of the plasma plume for estimating in-flight particle characteristics in APS. We introduce three different video-derived feature representations and evaluate them using Tabular Prior-Data Fitted Networks (TabPFN), convolutional neural networks (CNN), and classical regression baselines including Random Forest, Gradient Boosting, Support Vector Regression, and XGBoost. Experiments are conducted using grouped leave-one-out cross-validation on 126 labeled pre- and post-spray video recordings from 63 APS spray runs. Across the engineered feature experiments, TabPFN achieves the most consistent performance for temperature prediction, reaching R2 = 0.86 using the combined feature representation. CNN models particularly perform stronger for velocity prediction, achieving R2 of 0.81. In addition, we evaluate models operating directly on raw video frames using pretrained CNNs and find that the highest performance is achieved by a pretrained CNN with a regression head with R2 of 0.90 and 0.82 for temperature and velocity, respectively. The results demonstrate that video-derived plume information provides a promising and scalable foundation for non-invasive APS diagnostics and real-time process monitoring.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript investigates using high-speed video of the plasma plume in atmospheric plasma spraying (APS) to predict in-flight particle temperature and velocity. Three video-derived feature sets and raw frames are evaluated with TabPFN, CNNs, and classical regressors (RF, GB, SVR, XGBoost) under grouped leave-one-out cross-validation on 126 recordings from 63 APS runs. Reported peak performance is R²=0.86 (TabPFN, combined features, temperature), R²=0.90 (pretrained CNN, temperature) and R²=0.82 (pretrained CNN, velocity). The central claim is that video plume data provides a scalable, non-invasive foundation for real-time APS diagnostics.

Significance. If the predictive mappings generalize beyond the 63 runs, the work would supply a practical, camera-based alternative to invasive particle diagnostics, supporting process monitoring and quality control in thermal spray applications. The grouped CV design and direct comparison of engineered features versus raw-frame CNNs are methodologically sound elements that strengthen the contribution if the data envelope is adequately described.

major comments (3)
  1. [Experiments / Data] Data collection / Experiments: The 63 spray runs supplying the 126 recordings are never characterized with respect to the operating-parameter envelope (powder size distribution, carrier-gas flow, current, standoff distance, substrate conditions). Without this information the claim that the approach supplies a 'scalable foundation' for production monitoring cannot be evaluated; the grouped LOO CV result may simply reflect interpolation within a narrow slice of parameter space.
  2. [Methods] Methods: No description is given of how the ground-truth temperature and velocity labels were obtained (sensor model, calibration procedure, temporal alignment with video frames) nor of the precise definitions and extraction pipelines for the three 'engineered feature representations'. These omissions make the reported R² values impossible to reproduce or stress-test.
  3. [Results] Results: Although grouped LOO CV is used, the manuscript reports neither per-fold standard deviations, confidence intervals on the R² values, nor any sensitivity analysis to the particular grouping or to the 63-run sample. On a dataset of this size such statistics are required to substantiate the headline performance numbers.
minor comments (2)
  1. [Abstract / Methods] The abstract states 'pre- and post-spray video recordings' but the manuscript never clarifies what post-spray footage contributes to the particle-characteristic prediction task.
  2. [Methods] Notation for the three feature sets is introduced without an accompanying table or equation block that would allow a reader to map names to concrete image-processing steps.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and constructive feedback on our manuscript. We address each of the major comments below, indicating the revisions we plan to make to strengthen the paper.

read point-by-point responses

  1. Referee: [Experiments / Data] Data collection / Experiments: The 63 spray runs supplying the 126 recordings are never characterized with respect to the operating-parameter envelope (powder size distribution, carrier-gas flow, current, standoff distance, substrate conditions). Without this information the claim that the approach supplies a 'scalable foundation' for production monitoring cannot be evaluated; the grouped LOO CV result may simply reflect interpolation within a narrow slice of parameter space.

    Authors: We agree that characterizing the operating-parameter envelope is essential for evaluating the scalability of the proposed method. In the revised version of the manuscript, we will add a detailed section or table describing the ranges and specific values of the operating parameters used across the 63 spray runs, including powder size distribution, carrier-gas flow, current, standoff distance, and substrate conditions. This will provide context for the dataset and help assess whether the results represent interpolation within a limited parameter space or broader applicability. revision: yes

  2. Referee: [Methods] Methods: No description is given of how the ground-truth temperature and velocity labels were obtained (sensor model, calibration procedure, temporal alignment with video frames) nor of the precise definitions and extraction pipelines for the three 'engineered feature representations'. These omissions make the reported R² values impossible to reproduce or stress-test.

    Authors: We acknowledge the need for greater methodological detail to ensure reproducibility. We will revise the Methods section to include a full description of the sensor model and calibration procedure used to obtain the ground-truth temperature and velocity measurements, as well as the temporal alignment process with the video frames. Furthermore, we will provide precise definitions and step-by-step extraction pipelines for each of the three engineered feature representations. revision: yes

  3. Referee: [Results] Results: Although grouped LOO CV is used, the manuscript reports neither per-fold standard deviations, confidence intervals on the R² values, nor any sensitivity analysis to the particular grouping or to the 63-run sample. On a dataset of this size such statistics are required to substantiate the headline performance numbers.

    Authors: We concur that additional statistical reporting is necessary given the dataset size. In the revised manuscript, we will report per-fold standard deviations and confidence intervals for the R² values. We will also include a sensitivity analysis examining the impact of the grouping strategy and the sample of 63 runs on the performance metrics. revision: yes

Circularity Check

0 steps flagged

No circularity: standard supervised regression on held-out runs

full rationale

The paper reports empirical performance of TabPFN, CNNs, and classical regressors on video-derived features for temperature and velocity prediction. Evaluation uses grouped leave-one-out cross-validation on 126 recordings from 63 runs. No equations, derivations, fitted parameters renamed as predictions, or self-citation chains appear in the abstract or described methodology. All reported R² values are direct outputs of standard ML training and testing procedures with no reduction to inputs by construction. The work is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard supervised learning assumptions plus the empirical claim that video contains sufficient signal; no new physical entities or ad-hoc constants are introduced.

axioms (2)
  • domain assumption Labeled particle temperature and velocity measurements obtained during the 63 spray runs are accurate ground truth.
    The paper trains regression models against these labels; any systematic error in label acquisition would propagate directly into reported R² values.
  • domain assumption The 126 recordings are statistically representative of future APS runs under similar nominal conditions.
    Grouped leave-one-out CV assumes the held-out runs capture the relevant distribution; violation would invalidate generalization claims.

pith-pipeline@v0.9.1-grok · 5818 in / 1481 out tokens · 23760 ms · 2026-06-27T22:45:51.532772+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 1 canonical work pages

  1. [1]

    Journal of Thermal Spray Technology30, 1751–1764 (2021)

    Bobzin, K., Wietheger, W., Heinemann, H., Dokhanchi, S.R., Rom, M., Visconti, G.: Prediction of particle properties in plasma spraying based on machine learning. Journal of Thermal Spray Technology30, 1751–1764 (2021)

    2021

  2. [2]

    Journal of Thermal Spray Technology30(4), 987–1000 (2021) Video-Based Prediction of In-Flight Particle Characteristics in APS 17

    Bobzin, K., Öte, M., Knoch, M., Alkhasli, I., Heinemann, H.: High-speed video analysis of the process stability in plasma spraying. Journal of Thermal Spray Technology30(4), 987–1000 (2021) Video-Based Prediction of In-Flight Particle Characteristics in APS 17

    2021

  3. [3]

    Journal of Manufacturing Processes134, 1057–1068 (2025)

    Bokade, R., Müftü, S., Özdemir, O.Ç., Jin, X.: Thermal imaging based non- destructive testing for fault detection in cold spray additive manufacturing. Journal of Manufacturing Processes134, 1057–1068 (2025)

    2025

  4. [4]

    Machine Learning45(1), 5–32 (2001)

    Breiman, L.: Random forests. Machine Learning45(1), 5–32 (2001)

    2001

  5. [5]

    In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD)

    Chen, T., Guestrin, C.: Xgboost: A scalable tree boosting system. In: Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD). pp. 785–794 (2016)

    2016

  6. [6]

    Plasma Chemistry and Plasma Processing33(5), 993–1023 (2013)

    Choudhury, T., Berndt, C., Man, Z.: An extreme learning machine algorithm to predict the in-flight particle characteristics of an atmospheric plasma spray process. Plasma Chemistry and Plasma Processing33(5), 993–1023 (2013)

    2013

  7. [7]

    IEEE Transactions on Industrial Informatics (2025)

    Citarella, A.A., Carrino, L., De Marco, F., Di Biasi, L., Perna, A.S., Viscusi, A., Tortora, G.: Ai data-driven optimization of cold spray coating manufacturing. IEEE Transactions on Industrial Informatics (2025)

    2025

  8. [8]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) pp

    Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: Imagenet: A large-scale hierarchical image database. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) pp. 248–255 (2009)

    2009

  9. [9]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)

    Donahue, J., Hendricks, L.A., Guadarrama, S., Rohrbach, M., Venugopalan, S., Saenko, K., Darrell, T.: Long-term recurrent convolutional networks for visual recognition and description. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 2625–2634 (2015)

    2015

  10. [10]

    International Jour- nal of Computer Vision51(2), 91–109 (2003)

    Doretto, G., Soatto, S., Wu, Y., Chiuso, A.: Dynamic textures. International Jour- nal of Computer Vision51(2), 91–109 (2003)

    2003

  11. [11]

    Advances in Neural Information Processing Systems9(1997)

    Drucker, H., Burges, C.J.C., Kaufman, L., Smola, A., Vapnik, V.: Support vector regression machines. Advances in Neural Information Processing Systems9(1997)

    1997

  12. [12]

    Chapman & Hall/CRC (1993)

    Efron, B., Tibshirani, R.J.: An Introduction to the Bootstrap. Chapman & Hall/CRC (1993)

    1993

  13. [13]

    Annals of Statistics29(5), 1189–1232 (2001)

    Friedman, J.H.: Greedy function approximation: A gradient boosting machine. Annals of Statistics29(5), 1189–1232 (2001)

    2001

  14. [14]

    Journal of Thermal Spray Technology12, 44–52 (2003)

    Friis, M., Persson, C.: Control of thermal spray processes by means of process maps and process windows. Journal of Thermal Spray Technology12, 44–52 (2003)

    2003

  15. [15]

    Journal of adhesion science and technology18(4), 495–505 (2004)

    Guessasma, S., Montavon, G., Coddet, C.: Analysis of the influence of atmospheric plasma spray (aps) parameters on adhesion properties of alumina-titania coatings. Journal of adhesion science and technology18(4), 495–505 (2004)

    2004

  16. [16]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 770–778 (2016)

    2016

  17. [17]

    Neural Computation 9(8), 1735–1780 (1997)

    Hochreiter, S., Schmidhuber, J.: Long short-term memory. Neural Computation 9(8), 1735–1780 (1997)

    1997

  18. [18]

    Nature637, 319–326 (2025)

    Hollmann, N., et al.: Accurate predictions on small data with a tabular foundation model. Nature637, 319–326 (2025)

    2025

  19. [19]

    Hudomalj, U.: Monitoring, Modelling and Control for Increasing Repeatability of ThermalSpraying.Ph.D.thesis,ETHZurich(2024).https://doi.org/10.3929/ethz- b-000673305

    work page doi:10.3929/ethz- 2024

  20. [20]

    Journal of Mathematical Imaging and Vision60(9), 1369–1398 (Jun 2018)

    Jansson, Y., Lindeberg, T.: Dynamic texture recognition using time-causal and time-recursive spatio-temporal receptive fields. Journal of Mathematical Imaging and Vision60(9), 1369–1398 (Jun 2018)

    2018

  21. [21]

    desired coating structural attributes

    Kanta, A.F., Montavon, G., Planche, M.P., Coddet, C.: Artificial neural networks implementation in plasma spray process: Prediction of power parameters and in- flight particle characteristics vs. desired coating structural attributes. Surface and Coatings Technology203, 3361–3369 (2009) 18 A. Praveen et al

    2009

  22. [22]

    Expert systems with applications38(1), 260–271 (2011)

    Kanta, A.F., Montavon, G., Berndt, C.C., Planche, M.P., Coddet, C.: Intelligent system for prediction and control: Application in plasma spray process. Expert systems with applications38(1), 260–271 (2011)

    2011

  23. [23]

    Journal of Thermal Spray Technology11, 380–386 (2002)

    Leblanc, L., Moreau, C.: The long-term stability of plasma spraying. Journal of Thermal Spray Technology11, 380–386 (2002)

    2002

  24. [24]

    John Wiley & Sons (2011)

    Maggio, E., Cavallaro, A.: Video Tracking: Theory and Practice. John Wiley & Sons (2011)

    2011

  25. [25]

    Journal of Thermal Spray Technology33, 17–32 (2024)

    Malamousi, K., Delibasis, K., Kamnis, S.: Real-time thermal spray process mon- itoring using convolution neural network deep learning architectures. Journal of Thermal Spray Technology33, 17–32 (2024)

    2024

  26. [26]

    In: Proceedings of the International Thermal Spray Conference

    Moreau, C.: Towards a better control of thermal spray processes. In: Proceedings of the International Thermal Spray Conference. vol. 2, pp. 1681–1693 (1998)

    1998

  27. [27]

    In: International Conference on Learning Representations (2022)

    Müller, S., Hollmann, N., Arango, S.P., Grabocka, J., Hutter, F.: Transformers can do bayesian inference. In: International Conference on Learning Representations (2022)

    2022

  28. [28]

    (eds.): Atmospheric Pressure Plasma: From Diagnostics to Applications

    Nikiforov, A., Chen, Z. (eds.): Atmospheric Pressure Plasma: From Diagnostics to Applications. CRC Press, Boca Raton, FL, USA (2019)

    2019

  29. [29]

    Progress in Additive Manufacturing10, 4327– 4347 (2025)

    Prasad, C.D., Tiwari, A., Suryawanshi, S.R., Dileep, B.P., Gowda, A.C., Masum, H., Dutt, K.M., Sunil Prashanth Kumar, S., Bavan, S.: Overview of thermal spray coating on additive manufacturing. Progress in Additive Manufacturing10, 4327– 4347 (2025)

    2025

  30. [30]

    Savangouder, R.V., Patra, J.C., Palanisamy, S.: A machine learning technique for prediction of cold spray additive manufacturing input process parameters to achieveadesiredspraydepositprofile.IEEETransactionsonIndustrialInformatics 20, 12275–12283 (2024)

    2024

  31. [31]

    Com- putational Materials Science50, 805–809 (2010)

    Sha, W.: Comment on ‘modelling of the aps plasma spray process using artificial neural networks: Basis, requirements and an example’ by guessasma et al. Com- putational Materials Science50, 805–809 (2010)

    2010

  32. [32]

    Materials and Corrosion58(2), 92–102 (2007)

    Singh, H., Sidhu, B., Puri, D., Prakash, S.: Use of plasma spray technology for deposition of high temperature oxidation/corrosion resistant coatings–a review. Materials and Corrosion58(2), 92–102 (2007)

    2007

  33. [33]

    Journal of Thermal Spray Technology10, 94–104 (2001)

    Vattulainen, J., Hämäläinen, E., Hernberg, R., Vuoristo, P., Mäntylä, T.: Novel method for in-flight particle temperature and velocity measurements in plasma spraying using a single ccd camera. Journal of Thermal Spray Technology10, 94–104 (2001)

    2001

  34. [34]

    Journal of Thermal Spray Technology32, 175–187 (2023)

    Yu, K.R., Cojocaru, C.V., Ilinca, F., Irissou, E.: Ensemble methods for aps in-flight particle temperature and velocity prediction considering torch electrodes ageing. Journal of Thermal Spray Technology32, 175–187 (2023)

    2023

  35. [35]

    Surface and Coatings Technology394, 125862 (2020)

    Zhu, J., Wang, X., Kou, L., Zheng, L., Zhang, H.: Prediction of control parameters corresponding to in-flight particles in atmospheric plasma spray employing convo- lutional neural networks. Surface and Coatings Technology394, 125862 (2020)

    2020

  36. [36]

    Applied Surface Science563, 150098 (2021)

    Zhu, J., Wang, X., Kou, L., Zheng, L., Zhang, H.: Application of combined transfer learning and convolutional neural networks to optimize plasma spraying. Applied Surface Science563, 150098 (2021)

    2021

  37. [37]

    In: Graphics Gems IV, pp

    Zuiderveld, K.: Contrast limited adaptive histogram equalization. In: Graphics Gems IV, pp. 474–485. Academic Press (1994)

    1994